Pith. sign in

REVIEW 1 major objections 5 minor 43 references

WarpMPC: Large-Batch MPC on GPU via ADMM with Unrolled $LDL^\top$ Factorization

T0 review · 1 major / 5 minor · reviewed 2026-07-14 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Unrolling the shared sparse factorizations of identical-structure MPC batches lets a single GPU run 8,000–250,000 SQP iterations per second.

desk verdict Solid systems paper: fixed-pattern unrolled LDLᵀ plus three concrete GPU optimizations deliver real large-batch constrained-MPC throughput, with ablations and a hardware distillation demo that match the claim. read the letter →

arxiv 2607.11603 v1 pith:6OVEHRYC submitted 2026-07-13 cs.RO cs.SYeess.SYmath.OC

Henrik Hose , Se Hwan Jeon , Charles Khazoom , Sangbae Kim , Sebastian Trimpe This is my paper · ORCID
classification cs.ROcs.SYeess.SYmath.OC
keywords modelpredictivecontrolsequentialquadraticprogrammingADMMsparseLDLfactorizationGPUbatchingroboticsJAXsensitivities
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Model-predictive control is useful for robots that must respect hard constraints, but learning pipelines that need tens or hundreds of thousands of MPC solves have been forced either to run them one-by-one on CPUs or to rent large clusters. This paper shows that when every problem in a batch has exactly the same sparsity pattern—the same horizon, the same stage structure, the same cost and constraint nonzeros—the expensive sparse factorizations and backsolves that dominate an ADMM quadratic-program solver can be completely unrolled into fixed GPU kernels. Three further engineering choices (a symbolic-major memory layout, column segmentation that cuts padding waste, and dependency-level scheduling of the triangular solves) remove the remaining memory and launch bottlenecks. The resulting toolbox therefore solves large batches of nonlinear cartpole, quadrotor and humanoid whole-body problems at throughputs three to twenty-five times higher than the best competing GPU solvers, and can generate a million-scale training set for a neural-network MPC in a few minutes.

What carries the argument

Unrolled fixed-pattern LDLᵀ factorization of the ADMM KKT matrix: offline, the elimination tree, nonzero pattern of L and dependency levels are precomputed once; online, every batch element simply executes the same arithmetic on its own numerical values, with segmented kernels and level-scheduled triangular solves removing padding and sequential bottlenecks.

What would settle it

Measure end-to-end SQP throughput on the same cartpole, quadrotor and humanoid problems at batch sizes above 10 000; if the unrolled kernels no longer outperform cuDSS, PCG and dense baselines by the claimed 3–25 imes factor, or if they run out of memory sooner, the central claim fails.

Watch

Extended reading notes

Core claim

Because every MPC instance inside a large batch shares an identical sparsity pattern, a single offline symbolic LDLᵀ factorization can be unrolled into custom GPU kernels whose only online work is the numerical arithmetic of that fixed pattern; with a batched memory layout, padding-reducing column segmentation, and level-scheduled backsolves, those kernels deliver 8 000–250 000 SQP iterations per second on realistic nonlinear robot problems—three to twenty-five times the throughput of previous GPU baselines.

Load-bearing premise

Every problem inside the batch must share exactly the same sparsity pattern—identical horizon, stage structure, and nonzero locations—so that one offline symbolic factorization can be reused for all of them.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

1 major / 5 minor

Summary. The paper presents WarpMPC, a JAX/Warp toolbox for high-throughput batched nonlinear MPC on GPU. It solves large batches (10k–100k+) of SQP iterations that share a fixed sparsity pattern by unrolling sparse LDL^T factorizations and backsolves inside an OSQP-style ADMM QP solver. Three concrete optimizations—symbolic-major memory layout, dynamic-programming segmentation of factorization columns to reduce padding, and dependency-level scheduling of triangular solves—are introduced and ablated. Stage-wise SQP data are generated from CasADi via a custom string-based translation to JAX. Benchmarks on cartpole, quadrotor, and humanoid whole-body MPC report 8k–250k SQP iterations per second and 3–25× gains over PCG, MPAX, TurboMPC/cuDSS, and CPU OSQP; a Crazyflie hardware demo shows that a neural-network approximation distilled from the solver in under four minutes stabilizes the vehicle.

Significance. Large-batch constrained MPC is a practical bottleneck for imitation learning, residual RL, and dataset generation for approximate controllers. Existing GPU solvers either drop hard inequalities, rely on iterative linear algebra that loses accuracy, or fail to scale past a few thousand instances. WarpMPC supplies an open-source, fixed-pattern ADMM implementation that demonstrably reaches tens to hundreds of thousands of SQP iterations per second on realistic robotics problems while retaining hard constraints and reverse-mode sensitivities. The ablations, multi-baseline comparisons, and closed-loop hardware result make the contribution immediately usable and falsifiable. The open-source release and automatic C-code export for the distilled network further raise the work’s practical value.

major comments (1)
  1. The central throughput claim is measured under the explicit premise that every batch element shares an identical sparsity pattern (fixed horizon, stage structure, cost/constraint nonzeros). This premise is stated clearly in the abstract, §III and Algorithm 1 and is the natural regime for the intended learning pipelines; the reported numbers are therefore well-supported under the stated conditions. No load-bearing technical inconsistency appears that would overturn the claim.
minor comments (5)
  1. Table II lists humanoid horizon N=28 while the caption of Fig. 1 and the text of §IV-D use N=27; a single consistent value should be used throughout.
  2. Fig. 7 bottom row reports mean constraint violations for LDL^T versus PCG; adding the corresponding numbers for MPAX and TurboMPC (where they complete) would make the accuracy comparison complete.
  3. The abstract and introduction claim “under 4 minutes” for dataset synthesis plus training, while §V reports 2 min synthesis + 1.7 min training + 11 min compilation that “can be cached.” Clarifying that the 4-minute figure excludes one-time compilation would avoid ambiguity.
  4. A short sentence in §III-B noting that the segment-budget DP is solved once offline and that the chosen budgets (384/256/96) are those used for all subsequent timing would help readers reproduce the exact kernels.
  5. Minor typos: “optimziations,” “usefullness,” “significanlty,” “enviornments,” and “infill” (for fill-in) appear in the conclusion and figure captions.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical systems paper whose throughput claims are measured against external baselines under an explicitly stated fixed-sparsity premise.

full rationale

WarpMPC is an engineering/systems contribution: it unrolls a standard sparse LDLᵀ factorization of the fixed-pattern OSQP KKT matrix (Eqs. 6, 11–15), adds three implementation optimizations (symbolic-major layout, optimal column segmentation via DP on Eq. 17, level-scheduled triangular solves), and reports measured SQP-iteration throughput on cartpole, quadrotor and humanoid benchmarks against independent GPU/CPU baselines (cuDSS, PCG, MPAX, TurboMPC, dense BoxOSQP, single-core OSQP). The central premise—that every instance in a batch shares identical sparsity—is stated up front (abstract, §III, Algorithm 1) and is the regime actually measured; it is not smuggled in as a derived result. No quantity is fitted to data and later presented as a prediction, no uniqueness theorem is imported from the authors’ prior work to forbid alternatives, and self-citations (e.g., to the humanoid formulation of Khazoom et al. or the OSQP sensitivity formulas) serve only as problem definitions or standard algorithmic building blocks, not as load-bearing premises that force the throughput numbers. The derivation chain is therefore self-contained and non-circular; the honest finding is score 0.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

As a systems/optimization paper the load-bearing premises are standard numerical linear algebra plus the domain assumption that every batch member shares an identical sparsity pattern. Free parameters are ordinary solver hyper-parameters (segment budget, ADMM iteration counts, proximal σ, ρ) chosen by the authors for the reported experiments; no new physical entities are postulated.

free parameters (2)
  • segment budget n_seg
    Chosen offline by dynamic programming (Eq. 17) and set per benchmark (Table II: 384/256/96); directly controls padding waste and kernel-launch overhead.
  • ADMM iteration counts and proximal parameters (σ, ρ, ω_rel)
    Fixed a priori (Table II: 25–100 QP iterations) rather than adapted online; affect both accuracy and measured throughput.
assumptions (3)
  • domain assumption All problems in a batch share identical sparsity of the KKT matrix (fixed horizon, stage structure, cost and constraint sparsity).
    Stated in §III and Algorithm 1; enables a single offline symbolic factorization and unrolled kernels.
  • standard math Sparse LDL^T factorization of a quasi-definite KKT matrix is numerically stable for the chosen proximal regularization σ > 0.
    Classic result used without re-proof (Stewart 2003, OSQP literature).
  • domain assumption Gauss-Newton Hessian approximation remains positive-semidefinite for the SQP subproblems.
    Standard SQP practice for MPC; no second-order corrections are performed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of WarpMPC: Large-Batch MPC on GPU via ADMM with Unrolled $LDL^\top$ Factorization." pith.science (2026). https://pith.science/paper/6OVEHRYC

@misc{pith2026260711603,
  author       = {Pith},
  title        = {Pith review of: WarpMPC: Large-Batch MPC on GPU via ADMM with Unrolled $LDL^\top$ Factorization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6OVEHRYC}},
  note         = {Machine review of arXiv:2607.11603}
}
abstract

This paper introduces numerical optimizations for maximizing throughput on GPU when solving large batches (10,000 to over 100,000) of sequential quadratic programming (SQP) iterations, where all problems have the same structure. The optimizations are implemented in a toolbox WarpMPC for model-predictive control (MPC) in JAX and Warp. Based on the insight that all MPC problem instances in a batch share the same sparsity in time, cost, and constraints, we propose unrolling sparse linear factorizations and solves, which dominate alternating direction method of multipliers (ADMM) solver runtime. We avoid memory access bottlenecks and wasting computations via optimized memory layout, padding-reducing segmentation of the unrolled factorization, and dependency level scheduled backsolves, additionally accelerating sensitivity computation. We achieve throughputs of 8,000 to 250,000 SQP iterations per second on nonlinear cartpole, quadrotor, and humanoid robot benchmarks, outperforming baselines by 3$\times$ to 25$\times$. We illustrate practical usefulness by synthesizing a dataset and training a neural network approximation of an MPC in under 4 minutes that stabilizes a nano quadrotor in hardware experiments.

Figures

Figures reproduced from arXiv: 2607.11603 by the authors.

Figure 1
Figure 1. Sparsity patterns of the full KKT systems [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Visualization of optimal column segmentation for the cartpole, [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. Code export, compilation, and execution time over number of floating [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (7 more)
Figure 6
Figure 6. Figure 6: Throughput in QPs per second over batch size for a linear MPC [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 5
Figure 5. Figure 5: Effect of optimizations for a single sparse factorization (left) and [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 9
Figure 9. Figure 9: Improvement of maximum throughput using the optimized WarpMPC [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: Hardware experiments controlling a Crazyflie nano quadrotor [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]
Figure 11
Figure 11. Figure 11: For WarpMPC, we report the constraint satisfaction [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]
Figure 8
Figure 8. Figure 8: Humanoid whole-body MPC with forward (left) and sideways (center) [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 11
Figure 11. Figure 11: Constraint satisfaction and throughput in simulation steps per second [PITH_FULL_IMAGE:figures/full_fig_p008_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 5 linked inside Pith

  1. [1]

    J. B. Rawlings, D. Q. Mayne, and M. M. Diehl,Model Predictive Control: Theory, Computation, and Design. Nob Hill Publishing, 2017

    2017

  2. [2]

    Actor-critic model predictive control,

    A. Romero, Y . Song, and D. Scaramuzza, “Actor-critic model predictive control,” inICRA. IEEE, 2024, pp. 14 777–14 784

    2024

  3. [3]

    Residual MPC: Blending reinforcement learning with GPU-parallelized model predictive control,

    S. H. Jeon, H. J. Lee, S. Hong, and S. Kim, “Residual MPC: Blending reinforcement learning with GPU-parallelized model predictive control,” arXiv preprint arXiv:2510.12717, 2025

    arXiv 2025

  4. [4]

    Safe and fast tracking on a robot manipulator: Robust MPC and neural network control,

    J. Nubert, J. K ¨ohler, V . Berenz, F. Allg¨ower, and S. Trimpe, “Safe and fast tracking on a robot manipulator: Robust MPC and neural network control,”IEEE Rob. and Aut. Let., vol. 5, no. 2, pp. 3050–3057, 2020

    2020

  5. [5]

    Parameter-adaptive approximate MPC: Tuning neural-network controllers without retraining,

    H. Hose, A. Gr ¨afe, and S. Trimpe, “Parameter-adaptive approximate MPC: Tuning neural-network controllers without retraining,” in6th Annual Learning for Dynamics & Control Conference, 2024

    2024

  6. [6]

    acados—a modular open-source framework for fast embedded optimal control,

    R. Verschueren, G. Frison, D. Kouzoupis, J. Frey, N. v. Duijkeren, A. Zanelli, B. Novoselnik, T. Albin, R. Quirynen, and M. Diehl, “acados—a modular open-source framework for fast embedded optimal control,”Math. Prog. Comp., vol. 14, no. 1, pp. 147–183, 2022

    2022

  7. [7]

    OSQP: An operator splitting solver for quadratic programs,

    B. Stellato, G. Banjac, P. Goulart, A. Bemporad, and S. Boyd, “OSQP: An operator splitting solver for quadratic programs,” in2018 UKACC 12th international conference on control (CONTROL). IEEE, 2018, pp. 339–339

    2018

  8. [8]

    GPU acceleration of ADMM for large-scale quadratic programming,

    M. Schubiger, G. Banjac, and J. Lygeros, “GPU acceleration of ADMM for large-scale quadratic programming,”Journal of Parallel and Dis- tributed Computing, vol. 144, pp. 55–67, 2020

    2020

Show all 43 references
  1. [9]

    Exploiting GPU/SIMD architectures for solving linear-quadratic MPC problems,

    D. Cole, S. Shin, F. Pacaud, V . M. Zavala, and M. Anitescu, “Exploiting GPU/SIMD architectures for solving linear-quadratic MPC problems,” arXiv preprint arXiv:2209.13049, 2022

    2022 arXiv

  2. [10]

    MPCGPU: Real-time nonlinear model predictive control through preconditioned conjugate gradient on the GPU,

    E. Adabag, M. Atal, W. Gerard, and B. Plancher, “MPCGPU: Real-time nonlinear model predictive control through preconditioned conjugate gradient on the GPU,” inICRA. IEEE, 2024, pp. 9787–9794

    2024

  3. [11]

    GPU-accelerated dynamic nonlinear optimiza- tion with ExaModels and MadNLP,

    F. Pacaud and S. Shin, “GPU-accelerated dynamic nonlinear optimiza- tion with ExaModels and MadNLP,” in2024 IEEE 63rd Conference on Decision and Control (CDC). IEEE, 2024, pp. 5963–5968

    2024

  4. [12]

    Fast and certifiable trajectory optimization,

    S. Kang, X. Xu, J. Sarva, L. Liang, and H. Yang, “Fast and certifiable trajectory optimization,”arXiv preprint arXiv:2406.05846, 2024

    2024 arXiv

  5. [13]

    Differentiable MPC for end-to-end planning and control,

    B. Amos, I. D. Jimenez Rodriguez, J. Sacks, B. Boots, and J. Z. Kolter, “Differentiable MPC for end-to-end planning and control,” inAdv. in Neur. Inf. Proc. Sys., 2018

    2018

  6. [14]

    Trajax: Differen- tiable trajectory optimization with JAX,

    R. Frostig, M. Johnson, C. Leary, and D. Maclaurin, “Trajax: Differen- tiable trajectory optimization with JAX,” Google, Tech. Rep., 2021

    2021

  7. [15]

    Differentiable model predictive control on the GPU,

    E. Adabag, M. Greiff, J. Subosits, and T. J. Lew, “Differentiable model predictive control on the GPU,” inICLR, 2026

    2026

  8. [16]

    Primal-dual iLQR for GPU-accelerated learning and control in legged robots,

    L. Amatucci, J. Sousa-Pinto, G. Turrisi, D. Orban, V . Barasuol, and C. Semini, “Primal-dual iLQR for GPU-accelerated learning and control in legged robots,”IEEE Rob. and Aut. Let., vol. 11, no. 1, pp. 1010– 1017, 2025

    2025

  9. [17]

    DTC: Deep tracking control,

    F. Jenelten, J. He, F. Farshidian, and M. Hutter, “DTC: Deep tracking control,”Science Robotics, vol. 9, no. 86, p. eadh5401, 2024

    2024

  10. [18]

    Diffusion-based approximate MPC: Fast and consistent imitation of multi-modal action distributions,

    P. M. Julbe, J. Nubert, H. Hose, S. Trimpe, and K. J. Kuchenbecker, “Diffusion-based approximate MPC: Fast and consistent imitation of multi-modal action distributions,” inIROS. IEEE, 2025, pp. 5633– 5640

    2025

  11. [19]

    Nocedal and S

    J. Nocedal and S. J. Wright,Numerical optimization. Springer, 2006

    2006

  12. [20]

    GATO: GPU- accelerated and batched trajectory optimization for scalable edge model predictive control,

    A. Du, E. Adabag, G. Bravo-Palacios, and B. Plancher, “GATO: GPU- accelerated and batched trajectory optimization for scalable edge model predictive control,”arXiv preprint arXiv:2510.07625, 2025

    2025 arXiv

  13. [21]

    Building an old-fashioned sparse solver,

    G. Stewart, “Building an old-fashioned sparse solver,” University of Maryland, Tech. Rep., 2003

    2003

  14. [22]

    [Online]

    NVIDIA Corporation,cuDSS: A High-Performance CUDA Library for Direct Sparse Solvers, 2026, accessed: 2026-05-22. [Online]. Available: https://docs.nvidia.com/cuda/cudss/index.html

    2026

  15. [23]

    Linear solvers for power grid optimization problems: a review of GPU-accelerated linear solvers,

    K. ´Swirydowicz, E. Darve, W. Jones, J. Maack, S. Regev, M. A. Saunders, S. J. Thomas, and S. Pele ˇs, “Linear solvers for power grid optimization problems: a review of GPU-accelerated linear solvers,” Parallel Computing, vol. 111, p. 102870, 2022

    2022

  16. [24]

    TurboMPC: Fast, scalable, and differentiable model predictive control on the GPU,

    G. Bravo-Palacios, J. Zhang, Z. Pestrikov, B. Plancher, and T. Lew, “TurboMPC: Fast, scalable, and differentiable model predictive control on the GPU,”arXiv preprint arXiv: 2606.24039, 2026

    2026 arXiv

  17. [25]

    CusADi: A GPU parallelization framework for symbolic expressions and optimal control,

    S. H. Jeon, S. Hong, H. J. Lee, C. Khazoom, and S. Kim, “CusADi: A GPU parallelization framework for symbolic expressions and optimal control,”IEEE Rob. and Aut. Let., vol. 10, no. 2, pp. 899–906, 2024

    2024

  18. [26]

    CasADi – A software framework for nonlinear optimization and optimal control,

    J. A. E. Andersson, J. Gillis, G. Horn, J. B. Rawlings, and M. Diehl, “CasADi – A software framework for nonlinear optimization and optimal control,”Math. Prog. Comp., vol. 11, no. 1, pp. 1–36, 2019

    2019

  19. [27]

    W. E. Hart, C. D. Laird, J.-P. Watson, D. L. Woodruff, G. A. Hackebeil, B. L. Nicholson, J. D. Siirolaet al.,Pyomo-optimization modeling in python. Springer, 2017, vol. 67

    2017

  20. [28]

    do-MPC: Towards FAIR nonlinear and robust model predictive control,

    F. Fiedler, B. Karg, L. L ¨uken, D. Brandner, M. Heinlein, F. Brabender, and S. Lucia, “do-MPC: Towards FAIR nonlinear and robust model predictive control,”Contr. Eng. Pract., vol. 140, p. 105676, 2023

    2023

  21. [29]

    Learning for CasADi: Data-driven models in numerical optimization,

    T. Salzmann, J. Arrizabalaga, J. Andersson, M. Pavone, and M. Ryll, “Learning for CasADi: Data-driven models in numerical optimization,” in6th Annual Learning for Dynamics & Control Conference. PMLR, 2024, pp. 541–553

    2024

  22. [30]

    L4acados: Learning-based models for acados, applied to Gaussian process-based predictive control,

    A. Lahr, J. N ¨af, K. P. Wabersich, J. Frey, P. Siehl, A. Carron, M. Diehl, and M. N. Zeilinger, “L4acados: Learning-based models for acados, applied to Gaussian process-based predictive control,”IEEE Trans. on Contr. Sys. Tech., 2026

    2026

  23. [31]

    On the differentiability of the primal-dual interior-point method,

    K. Tracy and Z. Manchester, “On the differentiability of the primal-dual interior-point method,”arXiv preprint arXiv:2406.11749, 2024

    2024 arXiv

  24. [32]

    MPAX: Mathematical programming in JAX,

    H. Lu, Z. Peng, and J. Yang, “MPAX: Mathematical programming in JAX,” inNeurIPS Workshop on GPU-Accelerated and Scalable Optimization, 2025

    2025

  25. [33]

    HyKKT: a hybrid direct-iterative method for solving KKT linear systems,

    S. Regev, N.-Y . Chiang, E. Darve, C. G. Petra, M. A. Saunders, K. ´Swirydowicz, and S. Peleˇs, “HyKKT: a hybrid direct-iterative method for solving KKT linear systems,”Optimization Methods and Software, vol. 38, no. 2, pp. 332–355, 2023

    2023

  26. [34]

    ReLU-QP: A GPU-accelerated quadratic programming solver for model-predictive control,

    A. L. Bishop, J. Z. Zhang, S. Gurumurthy, K. Tracy, and Z. Manch- ester, “ReLU-QP: A GPU-accelerated quadratic programming solver for model-predictive control,” inICRA. IEEE, 2024, pp. 13 285–13 292

    2024

  27. [35]

    On the implementation of an interior- point filter line-search algorithm for large-scale nonlinear programming,

    A. W ¨achter and L. T. Biegler, “On the implementation of an interior- point filter line-search algorithm for large-scale nonlinear programming,” Mathematical programming, vol. 106, no. 1, pp. 25–57, 2006

    2006

  28. [36]

    Per- ceptive locomotion through nonlinear model-predictive control,

    R. Grandia, F. Jenelten, S. Yang, F. Farshidian, and M. Hutter, “Per- ceptive locomotion through nonlinear model-predictive control,”IEEE Transactions on Robotics, vol. 39, no. 5, pp. 3402–3421, 2023

    2023

  29. [37]

    JaxADi: Bridging CasADi and JAX for efficient numerical computing,

    I. Alentev, L. Kozlov, and S. Nedelchev, “JaxADi: Bridging CasADi and JAX for efficient numerical computing,” 2024. [Online]. Available: https://github.com/based-robotics/jaxadi

    2024

  30. [38]

    Efficient and modular implicit differentiation,

    M. Blondel, Q. Berthet, M. Cuturi, R. Frostig, S. Hoyer, F. Llinares- L´opez, F. Pedregosa, and J.-P. Vert, “Efficient and modular implicit differentiation,” inAdv. in Neur. Inf. Proc. Sys., 2022

    2022

  31. [39]

    An efficient real-time NMPC for quadrotor position control under communication time-delay,

    B. B. Carlos, T. Sartor, A. Zanelli, G. Frison, W. Burgard, M. Diehl, and G. Oriolo, “An efficient real-time NMPC for quadrotor position control under communication time-delay,” in16th Int. Conf. on Control, Automation, Robotics and Vision (ICARCV), 2020, pp. 982–989

    2020

  32. [40]

    Tai- loring solution accuracy for fast whole-body model predictive control of legged robots,

    C. Khazoom, S. Hong, M. Chignoli, E. Stanger-Jones, and S. Kim, “Tai- loring solution accuracy for fast whole-body model predictive control of legged robots,”IEEE Rob. and Aut. Let., vol. 9, no. 12, pp. 11 074– 11 081, 2024

    2024

  33. [41]

    An implicit time-stepping scheme for rigid body dynamics with coulomb friction,

    D. Stewart and J. C. Trinkle, “An implicit time-stepping scheme for rigid body dynamics with coulomb friction,” inICRA, vol. 1. IEEE, 2000, pp. 162–169

    2000

  34. [42]

    Learning an approximate model predictive controller with guarantees,

    M. Hertneck, J. K ¨ohler, S. Trimpe, and F. Allg ¨ower, “Learning an approximate model predictive controller with guarantees,”IEEE Contr. Sys. Let., vol. 2, no. 3, pp. 543–548, 2018

    2018

  35. [43]

    Deep learning-based model predictive control for resonant power converters,

    S. Lucia, D. Navarro, B. Karg, H. Sarnago, and O. Lucia, “Deep learning-based model predictive control for resonant power converters,” IEEE Trans. on Ind. Inf., vol. 17, no. 1, pp. 409–420, 2020

    2020

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.